Extracellular domains of E-cadherin determine key mechanical phenotypes of an epithelium through cell- and non-cell-autonomous outside-in signalling

Cadherins control intercellular adhesion in most metazoans. In vertebrates, intercellular adhesion differs considerably between cadherins of type-I and type-II, predominantly due to their different extracellular regions. Yet, intercellular adhesion critically depends on actomyosin contractility, in which the role of the cadherin extracellular region is unclear. Here, we dissect the roles of the Extracellular Cadherin (EC) Ig-like domains by expressing chimeric E-cadherin with E-cadherin and cadherin-7 Ig-like domains in cells naturally devoid of cadherins. Using cell-cell separation, cortical tension measurement, tissue-scale stretching and migration assays, we show that distinct EC repeats in the extracellular region of cadherins differentially modulate epithelial sheet integrity, cell-cell separation forces, and cell cortical tension through a Cdc42 pathway, which further differentially regulate epithelial tensile strength, ductility, and ultimately collective migration. Interestingly, dissipative processes rather than static adhesion energy mostly dominate cell-cell separation forces. We provide a framework for the emergence of epithelial phenotypes from cell mechanical properties dependent on EC outside-in signalling.

Cadherins control intercellular adhesion in most metazoans. In vertebrates, intercellular adhesion differs considerably between cadherins of type-I and type-II, predominantly due to their different extracellular regions. Yet, intercellular adhesion critically depends on actomyosin contractility, in which the role of the cadherin extracellular region is unclear. Here, we dissect the roles of the Extracellular Cadherin (EC) Ig-like domains by expressing chimeric E-cadherin with E-cadherin and cadherin-7 Ig-like domains in cells naturally devoid of cadherins. Using cell-cell separation, cortical tension measurement, tissue-scale stretching and migration assays, we show that distinct EC repeats in the extracellular region of cadherins differentially modulate epithelial sheet integrity, cell-cell separation forces, and cell cortical tension through a Cdc42 pathway, which further differentially regulate epithelial tensile strength, ductility, and ultimately collective migration. Interestingly, dissipative processes rather than static adhesion energy mostly dominate cell-cell separation forces. We provide a framework for the emergence of epithelial phenotypes from cell mechanical properties dependent on EC outside-in signalling.
Intercellular adhesion plays an important role in the development and maintenance of multicellular organisms [1][2][3] . Such adhesions are established and maintained through calcium-dependent, cell-cell adhesion molecules known as cadherins. Type-I and type-II cadherins, also known as classical cadherins, form one of six highly conserved branches of the cadherin superfamily, classified based on protein sequences in the extracellular cadherin (EC) domains. Classical cadherins typically have an extracellular region comprising five EC repeats (EC1-5), a single transmembrane region, and a cytoplasmic region 4,5 , and are often found concentrated in adherens junctions, dynamically interacting with the contractile cytoskeleton through catenin complexes. Despite striking structural similarities, E-cadherin (type-I) engenders robust intercellular adhesion, as seen in the epithelium, whereas cadherin-7 (type-II) is associated with much weaker adhesions, such as that in the mesenchyme 6,7 . This phenotypic difference may be attributed to the cadherin extracellular region. Indeed, whereas all three regions of cadherins are essential for normal function, the EC region dictates differences in cell-cell separation forces (SFs)-a metric that may be used as a proxy for intercellular adhesion energy 8 -between type-I and type-II cadherinexpressing cell pairs 6 .
In E-cadherin (E-cad) EC domains, henceforth referred to as EECs, the first and outermost EC domain (EEC1) has a tryptophan residue at position 2 (Trp2), which interacts with a hydrophobic pocket in the opposing EEC1 of its cadherin pair to form strand-swap dimers. In contrast, in cadherin-7 (cad-7) EC domains, henceforth referred to as 7ECs, the conserved Trp2 and Trp4 residues in 7EC1 anchor into a larger hydrophobic pocket to form two strand-swaps 9 . Associations and dissociations of strand-swapped dimers were recently identified to be mediated by binding intermediates referred to as X-dimers, which formed through extensive surface interactions between the residues near the EC1-2 calcium binding sites [10][11][12] . X-dimers behave as catch bonds, where the bond lifetime increases in the presence of tensile forces 13 ; comparatively, strand-swap dimers form slip bonds, for which the bond lifetime decreases under tensile force. Structural and biophysical studies corroborate that cadherins tune their trans-bond lifetimes by switching between X-and strandswap dimer states 14 . As both type-I and type-II cadherins are known to form Xdimers, whether the strand-swapping specifics of type-I and -II cadherins, or other causes are responsible for the significant difference between E-cad and cad-7 mediated cell-cell SFs remains to be uncovered 10,15 .
Indeed, although the residues involved in strand-swap and X dimerizations are exclusive to EC1,2 domains, mounting evidence suggests that the modular cadherin structure can also engage in multiple, trans-and cis-bonds involving more than the outer two EC Ig-like domains [16][17][18][19][20][21][22][23][24] . Moreover, in silico studies show that EEC and 7EC domains unfold in remarkably different ways under mechanical forces 25 . This raises the possibility that differences in adhesive properties between type-I and type-II cadherins arise from other ECs than EC1,2, and do not rely uniquely on strand-swap or X dimerizations.
Dual pipette cell-cell SF assays revealed that, after a few minutes of initial cell-cell contact, cell-cell SF is crucially dependent on the actin cytoskeletal anchorage of cadherins 26 , in a cadherin type-dependent manner 6 , and this is further supported by observations that cadherins uncouple from the cytoskeleton rather than from each other in zebrafish embryo cell-cell separation 27 . Indeed, cadherins appear to contribute to intercellular adhesion energy by locally down-regulating actomyosin cortical contractility [28][29][30] , with marginal contributions from the molecular binding energy of trans-interacting E-cad between cells 27,31,32 . Nevertheless, myosin-II-driven actin dynamics do modulate the immobilization of E-cad at the cell-cell contact periphery in a mechanosensitive manner 33 , where cadherin trans-interactions are required to physically link cells together 27,29 . Whether specific ECs of type-I and -II cadherins differentially regulate the cortical actomyosin cytoskeleton, and whether such a regulation depends on trans-interactions remain to be demonstrated.
In this work, we thus sought to address how the distinct ECs of type-I and -II cadherins regulate cell-cell adhesive properties, the contributions of trans-interactions and their effects on cortical mechanics, and the consequences at the tissue level. To do so, we examined the properties of cells individually expressing four different chimeric cadherins whose five EC regions are combinations of EC domains from Ecad or cad-7. Each chimera retained the E-cad transmembrane and cytoplasmic regions to preserve the interactions of E-cad, the catenin complex, and the cytoskeleton. We found that epithelial-like phenotype remained after exchanging EEC1-3 to 7EC1-3 but was fully lost upon exchange of the 4 first domains. Using a novel pillar-pipette assay to measure cell-cell SFs, we found that the exchange of the sole EEC3 by 7-EC3 lead to a very strong decrease of SF, and the replacement of the 3 domains EEC1-3 drastically reduced SF maturation. Next, we determined by pipette aspiration that EEC4 and EEC1-3 antagonistically affected CdC42 activity and cell cortical tension cell-autonomously. From this, and the measurement of cell-cell compaction, we determined that the static adhesion energy was a minor contributor to cell-cell SF, which was therefore mostly governed by dissipative processes to which EC region stretching before trans-bond rupture quantitatively suffices to contribute.
Finally, using a custom substrate-free tissue stretcher, we provide evidence that EEC3 offers tensile strength to cell sheets, while EEC1-3 provides ductility, in accordance with their respective effects on cell-cell SF and cortical tension. We propose that the antagonistic effects of tissue ductility and cohesion endowed by the different EC regions ultimately determines the speed of collective cell migration.

Cadherin/catenin complex recruitment and epithelial integrity is largely independent of EEC1-3 (but not of EEC1-4) domains .
To study the activities of the EEC domains, we designed chimeric mouse E-cad-cad-7 ectodomains -(N)77EEE(C), 777EE, EE7EE and 7777E-connected to the transmembrane and cytoplasmic regions of E-cad tagged with eGFP at the C-terminus ( Fig. 1a,b). Computer models of juxtaposed domains at the hybrid interfaces showed similar domain orientations between chimeras at each interface (e.g. compare 77EEE versus EE777) suggesting that the design of the chimeras did not disrupt domain:domain interactions. Furthermore, analysis of the calcium-binding site demonstrated that at least three sidechains were available to coordinate each calcium ion, supporting that the domain interfaces are stabilized by calcium in a similar manner to the native proteins (Fig. 1c) 34 . These chimeras and wild-type E-cad (EEEEE) were stably expressed individually in S180 cells using a lentivirus-based transduction technique, selected by Puromycin resistance and sorted using fluorescence activated cell-sorting (FACS) to pool eGFP positive cells ( Supplementary Fig. 1). S180 untransfected cells were used as a control.
Although the cells were sorted by FACS based on overall eGFP signal, cadherin surface levels are an important determinant of the maximum cell-cell SF and its maturation kinetics 6 . As we have used chimeric cadherin ectodomains in this study, the option of using antibodies targeting the ectodomains for labeling cell surface cadherins for FACS sorting did not seem feasible. Therefore, to determine the cell surface cadherin levels, we quantified eGFP intensity at the cell surfaces in contact with the substrate using confocal microscopy on fixed samples (Supplementary Fig. 2a-e). The intensities of cell-surface chimeric cadherins were normalized to E-cad intensity. The surface levels of all chimeric cadherins were significantly lower (P ≤ 0.04), but no more than 20%, than those of E-cad ( Fig. 1e and Supplementary Fig. 2ae). We also determined cadherin-eGFP intensity at cell-cell junctions in 2D culture by confocal microscopy on fixed samples, and found that junctional cadherin-eGFP intensities in 77EEE, 777EE and EE7EE clones were significantly higher than that of E-cad clones (P < 0.002; Fig. 1f). However, junctional levels of 7777E was significantly lower than that of all the other groups (P < 0.0001; Fig. 1f). Furthermore, the difference between the junctional levels of 7777E and E-cad was much larger than that between surface levels of the same group pair, or that between junctional levels of E-cad and any other chimeric cadherin. Together, these results show that the cadherin, catenin, and vinculin junctional recruitment hardly depends on EEC1-3. It suggests that the specifics of strand-swapping of type-I or -II cadherins may not be involved.
To verify the importance of strand-swapping, we stably expressed strand-swap incompetent, full-length E-cad mutants (W2A) tagged with eGFP in S180 cells using  (pre-existing doublets) in ultra-low adhesion dishes at a low cell density, were selected and attached to the side of a pillar tip, and a second doublet was made to adhere to the first doublet in series (Fig. 2a, b and Supplementary Fig. 3d). The cellcell junction between the two doublets matured over 1 hr; this is an optimal timescale for cells from all groups to form sufficiently strong cell-cell adhesions that cause pillar deflection during cell-cell separation while avoiding cell detachment from the pillar. The two doublets were separated from each other using a pipette by aspirating the free end of the 4-cell complex and manually moving the pipette along the y-axis and away from the pillar. This is performed using a micromanipulator at an average speed of ~3 µm/sec, while acquiring images at a rate of ~1 image per second to record the deflection in the pillar ( Fig. 2c and Supplementary Fig. 3d). Thereafter, the two cells of the pre-existing doublet still attached to the pillar are similarly separated ( junctions displayed a 2 fold reduction only. It suggests that EEC3 is essential for stronger SF short-term contacts. At longer time (18h) the SF strengthened for every chimeric cadherins. 777EE, EE7EE, 7777E only displayed a 2 fold reduction in SF (P = 0.17). 77EEE was indistinguishable from wild type E-cad (P < 0.0001). It suggested SF matured faster by strand-swapping in EEC1 than in 7EC1, provided that EEC3 is present. To confirm that the differences in SF observed here did not result from differences in cadherin surface levels, we normalized the SFs to cadherin surface levels. Normalized SFs at both time points followed the same trend as the raw SFs ( Supplementary Fig. 4a, b), thus data normalization to cadherin surface levels was considered dispensable from now on. In sum, the EEC3 is essential for strong cell-cell SFs, and together with EEC1,2 for fast strengthening.

EEC1-3 and EEC4 antagonistically modulate single cell cortical tension .
In the cell-cell SF assay, we noticed that cells expressing wild-type E-cad and 7777E showed larger deformations during separation as compared with 77EEE, 777EE and EE7EE (Supplementary Movie 2). Since cell cortical tension may affect cell deformation during this assay, and that intercellular adhesion strength depends on cell cortical tension 28,29 , we reasoned that different cadherin EC domains may have distinct effects on cortical tension.
To test this, we used the micropipette aspiration technique (Fig. 3a,b) to characterize the cortical tension of single cells and doublets derived from confluent 2D cultures by trypsin-less dissociation. The cortical tensions of cells expressing chimeric cadherins 77EEE, 777EE and EE7EE were significantly higher than that of cells expressing the wild-type E-cad or the chimeric 7777E ( Fig. 3c, P < 0.0001). The cortical tension of parental S180 cells (devoid of E-cad expression) was significantly lower than that of E-cad-expressing cells. Finally, the difference in cortical tension between E-cad and 7777E cells was much smaller than between these three groups and other chimeraexpressing cells. Strikingly, these results show that changing a single, specific EC domain in an intercellular adhesion protein appears to be sufficient to induce changes in single cell mechanical property.
To assess whether cortical tension also depends on intercellular adhesion, we then measured differences in cortical tension in doublets of E-cad-or 77EEE-expressing cells. We observed no significant difference between cortical tension of single cells and doublets of the same group (P = 0.07 and P = 0.5 respectively), and the difference in cortical tension between E-cad and 77EEE doublets was significant ( Fig.   3d, P < 0.0001). Therefore, cortical tension depends on the EC regions of expressed cadherins but not on intercellular adhesion.

Cadherin EC region modulates GTPase signalling to alter cortical tension
To investigate how cadherin EC region may affect cell cortical tension, we examined the activity of Cdc42 and Rac1 using a G-LISA assay on cell lysates. Indeed, distinct Rac1 activity-both cell-autonomous and cadherin trans-bond-dependent-occurs in response to the expression of different cadherin proteins in the same cellular background 35 . When compared with E-cad-expressing cells, we found a substantial increases in the active Cdc42 levels in cells expressing 77EEE (P < 0.0001; Fig. 3e), 777EE (P < 0.0001), EE7EE (P < 0.0001) . Cells expressing 7777E also exhibited an increase (P = 0.01), albeit much smaller. Active Cdc42 levels were also low in the parental S180 cells (P < 0.04). Active Rac1 levels of all the cell lines, however, were not significantly different from that of the E-cad expressing cells (P ≥ 0.1), with the exception of 7777E-expressing cells, which was lower (P = 0.049) (Fig. 3f). These results show that Cdc42 activity levels correlate with cell cortical tension, and therefore suggest that the EEC and 7EC regions may differentially regulate cortical tension through Cdc42.

Cell-cell separation forces are governed by dissipative processes
Within a previously established theoretical framework 8 , it is possible to assess the contribution of static intercellular adhesion energy to cell-cell SF. Indeed, the static intercellular adhesion energy W is defined as W = γ(1-cosθ), where γ is the cortical tension and cosθ the compaction parameter (Fig. 4a). In turn, the static SF, SF static = πRW, where R is the cell radius, in the limit of low compaction. Finally, the contact radius r is determined by r = Rsinθ. Thus, we set out to measure cell-cell compaction and the contact radius.
At 1 hr of contact, the trends of the mean contact radius and compaction parameter were seemingly indicative of the measured SF across different groups (Fig. 4b,d).
Indeed, E-cad-and 77EEE-expressing cells showed insignificant differences in contact radii and compaction parameters that were higher and lower, respectively, than that of the other chimeric EC cells. The SF, r and cosθ metrics appeared to behave consistently with higher intercellular adhesion energy in E-cad-and 77EEEcells compared with other chimeric EC cells. After 18 hrs, however, both the contact radius and compaction parameter narrowed to around 7 µm and 0.18, respectively, for all groups, without significant differences between groups ( Fig. 4 c, e).
We next computed the SF static from cortical tension and compaction parameters at 1 hr (where low compaction is a reasonable assumption) and 18 hrs. We found that SF static scaled well below the measured SFs in all conditions and, notably, at least an order of magnitude below that measured in cells expressing E-cad or 77EEE (Table 1). We conclude that static adhesion energy only marginally contributes to cell-cell SFs and that EEC1-3 contributions to cell-cell SFs essentially arise from dissipative processes.

EEC3 and EEC1-3 provide tensile strength and ductility, respectively, to epithelial cell sheets
Given the uncorrelated effects of the EC region on cell-cell SFs and cortical tension, we sought to assess how this combination would affect the mechanical properties of epithelial tissues. To avoid the confounding effects of cell-substrate adhesion, we developed a dispase-based epithelial tissue uniaxial stretcher that locally cleaves cell-substrate bonds, causing a local stretch on the adherent cell sheets in culture, and thereby shifting cell-generated tension entirely to cell-cell junctions ( Fig. 5a-d).
We found that, following dispase treatment, the E-cad cell sheet could withstand cellgenerated tension while thinning as a ductile material ( the inability of this chimera to support epithelial integrity (see above). Considering ductility as the inverse of the sheet cross-section area at rupture and tensile strength the pulling force at rupture, these results suggest that EEC3 alone provides tensile strength to the cohesive cell sheet, while together with EEC1,2 it provides ductility, so that the sheet can withstand larger strains before rupture Since the main phenotypic differences between wild-type and 77EEE cells appeared at the tissue level, in terms of ductility, we sought to determine the relevance of EC1,2-mediated dimerization in this process. As we had previously shown that epithelial integrity was impossible without strand-swapping (Supplementary movie 1, left panel), we stably expressed X-dimer incompetent (K14E mutation) full-length Ecad mutants tagged with eGFP in S180 cells by lentivirus-based transduction. In contrast with cells expressing strand-swap mutants, which bear no tensile-strength, cells expressing X-dimer-incompetent E-cad formed stable junctions comparable to those formed by cells expressing wild-type E-cad, and showed a ductile behavior similar to that of wild-type E-cad expressing cells (Supplementary movie 6). Thus, these results show that X-dimerization is neither essential nor accountable for differences in tissue ductility, pointing to a cell-autonomous mechanism, as for cell cortical tension.

EEC1-3 are essential for collective cell migration speed
A study proposed how cell-cell adhesion and cell cortical tension governed the dynamics of collective behaviours in adherent cell assemblies 36 . Thus, we next sought to assess how altering the material properties of cell sheets with different ECs would impact dynamic multicellular processes using wound-free gap closure assays.
Cells expressing 7777E showed the fastest gap closure compared with the other EC chimera-expressing cells (Fig. 6a-c and Supplementary Movie 7) but slower than Ecad expressing cells (P = 0.008), and this occurred as a non-cohesive cell assembly. This is consistent with epithelial integrity impeding cell migration speed.
Nevertheless, the rate of gap closure for E-cad-expressing cells was twice as fast as that of 77EEE-expressing cells (P < 0.0001); while the rate of 777EE-and EE7EEexpressing cells was significantly lower than that of E-cad (P < 0.0001 ). (Fig. 6c). As the EEC3 seems to aid faster migration and the lack of EEC4 abolishes cell cohesivity during migration, we wanted to check if just EEC3 and EEC4 are sufficient for fast and collective cell migration. We designed a 77EE7 chimera with the transmembrane and cytoplasmic domains from cad-7 and expressed in S180 cells. Owing to a low expression level, (Supplementary fig. 5a), the mean SF of 77EE7 doublets at 18 hrs of contact was also low (147 ± 58 nN) and comparable to the mean SF of 7777E expressing cells (P > 0.2) (Supplementary fig. 5b). However, despite low SF, the 77EE7 expressing cells migrated collectively and at a higher speed than the 777EE cells ( Supplementary Fig 5c). Together, these results are consistent with a positive effect of EEC3 on collective cell migration, which provides higher tensile strength, and of EEC1-3, which provides higher ductility, in a manner that compensates epithelial integrity effects caused by EEC4. Remarkably, this verifies previous predictions 36 .
A summary of the results is presented in Table 2.

Discussion
Cadherin expression is an important determinant of intercellular adhesive properties 6 .
Here we sought to characterize how Sthe EC domains of type-I and type-II cadherins affect cell-cell adhesion and tissue phenotypes. We expressed chimeras of E-cad and cad-7-preserving the calcium binding, transmembrane, and cytoplasmic regions of E-cad but swapping the various ectodomains of the two proteins-in cadherin-null cells, and sorted for medium overall expression based on eGFP. Although cadherin surface levels were not identical among the chimera-expressing cell groups (possibly due to small differences in expression levels), junctional cadherin levels could be modulated by the cadherin-type of EC domains. Indeed, EEC1-3 but not EEC1-4 were dispensable to accumulate cadherins at cell-cell contacts and recruit catenins and vinculin, thereby promoting cytoskeleton anchoring and adhesion strengthening 37 , 38 .
This shows that EC domains can regulate the recruitment of cadherin cytoplasmic partners through the control of cadherin incorporation at cell-cell contacts, in a cadherin-type specific manner. However, this cadherin-type specific recruitment of cytoplasmic partners does not rely on the strand-swap mechanism alone, although strand-swapping is required (as cells expressing strand-swap incompetent E-cad mutants didn't form stable junctions). Instead, our results are consistent with a role for EEC4 in this process, since EEC5 ectodomain is insufficient for epithelial integrity.
Interestingly, hypoglycosylation of EEC4, which occurs upon densification of cultures of epithelial cells, impacts the composition and stability of the cadherin/catenin complex, and impedes collective cell migration 39,40 . Moreover, normal glycosylation of EC4 of type-II VE-cadherin was previously found to prevent cadherin cis-clustering 41 . Future studies may investigate whether glycosylation-dependent cis-clustering and/or catenin recruitment mechanisms explains the difference between EC4s of type-I and -II cadherins in our experiments.
We used a pillar-pipette assay to quantify intercellular SFs, as this alleviates some of the potential limitations of the classical dual-pipette SF assay. In the dual-pipette assay, the cortical cytoskeleton is remodeled due to aspiration by the pipettes.
Moreover, the dual-pipette assay requires discrete increments in pipette aspiration, resulting in stress-relaxation cycles, which may reinforce the junction 42 , or change the cortical tension, and thus affect the measurements. Previous protocols for preparing pre-existing doublets from cell suspensions for SF assays were based the mechanical dissociation of confluent, 18 hrs cultures by pipetting 26 , but it is very rare to get pre-existing doublets in weakly adhering cell types as the cell-cell junctions often rupture during such mechanical dissociation process. Here, we used suspension cultures seeded at low density on ultra-low adhesion dishes to let cells form long-term doublets in the absence of cell-substrate adhesion by disrupting cell motility on substrates and thus minimizing the chances of doublets getting separated from each other after initial contact. Moreover, we previously observed that stimulating a cell of the doublet with a fibronectin coated bead for one hour promotes an increase in cell--cell SF 43 . The attachment of cell to the fibronectin coated pillar might also trigger such an increase in cell--cell separation force. These could be the reasons why we get higher SF values for pre-existing doublets as compared with results in previous studies (SF of E-cad: 638 ± 150 nN vs 350 nN).
Intercellular SFs between E-cad-expressing cells are much higher than that between cad-7 expressing cells, and this is mostly due to differences in extracellular domains 6 .
Here, our results show that SFs are high only in cells expressing cadherins bearing the EEC3. Previous studies on type-I N-cadherin support that EC3 glycosylation impedes cis-dimerization and thereby affects cadherin trans-bond kinetics 44,45 . As for EC4, a differential regulation of EC3 glycosylation-dependent clustering between E-cad and cad-7 may underlie the considerable differences in SFs between these cadherin types.
Previous structural and single-molecule studies have shown that both type-I and type-II cadherins are competent for both EC1,2-dependent strand-swap and X dimers, and proposed that the differences in the EC1 domain between the two types was sufficient to confer the type-specific activities of cadherins in terms of in vitro and in vivo cell sorting 9,10 . Here, strand-swap or X dimerization specifics of type-I and -II cadherins appear to only transiently affect SFs: SF strengthening being faster between EEC1,2 than between 7EC1-2, perhaps because strand-swapping kinetics differ. Indeed, this difference only arises when EEC3 is present, as if strand-swapping differences were EC3-dependent, reminiscently of EC3 glycosylation effects on trans-bond kinetics 45 .
In contrast, we show that mature SFs reach similar levels regardless of the type of EC1,2. Thus, differences in mature SFs are unlikely governed by differences in strand-swap or X-dimer properties between cadherin types, and SFs may not predict cell sorting or vice versa. In fact, our quantitative analysis of cell cortical tension and compaction revealed that the static intercellular adhesion energy-the quantity that does govern cell sorting 28,29 -only marginally contributes to cell-cell SFs. Therefore, we speculate that EEC3, possibly through glycosylation-dependent cis-clustering, contributes to dissipative mechanisms that allow cells to mechanically discriminate cell-cell separation from cell sorting. Consistently, we show that EEC3 provides tensile strength to epithelial sheets, which essentially relies on the ability of cells to bear stronger tensile forces before cell-cell contact rupture. Nevertheless, up to half the strength of mature separation forces remains EC type-independent. Thus, the cells appear to achieve with time a cadherin organization within the contact that is sufficient for a substantial contribution of dissipation upon cell-cell separation, regardless of the EC region specifics. This contribution may merely result from the cadherin tail/cytoskeleton driven accumulation of cadherins at the contact, since cadherin tail and cytoskeleton anchoring remains essential for SF strengthening, regardless of cadherin type 6,26 .
While EC-mediated cis-interactions and, at least partly independently, the cytoskeleton may synergize to accumulate stable cadherin complexes at cell-cell contacts, cell-cell separation ultimately involves the rupture of these cadherin- and Dock1, and result in a local dissolution of the actomyosin cortex 51,52 , which will in turn contribute to the intercellular adhesion energy and promote compaction 28,29 .
Interestingly, cadherin-trans-bonding, in a cadherin sub-type-dependent manner, regulates Rho GTPase activation and alter cell stiffness as measured by cell twisting cytometry 35 . This supports that cadherin regulation of Rho GTPases affects the actomyosin cortex at a distance from cell-cell contacts in a type-dependent manner.
Here, we show in addition that cells exhibit correlated Cdc42 activities and cortical tensions dependent on the cadherin EC domains but independent of intercellular adhesion. Therefore, cadherin EC domains, specifically EEC1-3 and EEC4, are involved in a cell-autonomous outside-in signaling targeting Rho GTPases and cortical tension. Despite the common involvement of EEC4 in cortical tension and epithelial integrity, the lack of correlation between these two properties as a function of EEC1-3 precludes that epithelial integrity emerges from high cell cortical tension.
In contrast, the tissue stretching assay reveals a correlation between low cortical tension and tissue ductility, which supports a model by which tissue ductility arises from cell-autonomous low cortical tension. Along the same line, cells expressing Ecads competent or incompetent for X-dimerization show the same ductile behavior as a tissue, further supporting that cortical tension and ductility-provided that stable cell-cell junctions can occur-are regulated independently of cadherin trans-bond regulation, and are therefore cell-autonomous.
A previous theoretical study had showed how cell-autonomous mechanical properties could determine collective behavior of epithelia, specifically implicating cell cortical tension, increase in which impedes epithelial remodeling, akin to a jamming transition 36 . Here, wound-healing assays reveal an apparent complex dependence on cadherin EC domains. Nevertheless, the effects of EC domains on intercellular contact formation, SFs, cortical tension, tissue tensile strength, and tissue ductility allow us to propose the following model. First, the formation of the cell layer devoid of intercellular gaps, supported by EEC4, dampens single-cell migration speed, and prevents the cohesive sheet from migrating as fast as a collection of individual cells, reminiscently of a density-driven jamming 36 . Second, the EEC3 provides tissue tensile strength, which allows cells to pull harder on each other while migrating, thereby promoting collective migration speed. Note that we had previously demonstrated that cadherin-mediated adhesion stimulates cell spreading and that stronger cell-cell adhesion results in stronger traction forces at focal adhesions through the Src and P13K signaling pathways 53 . This mechanism could also contribute to faster collective cell migration speeds in cells with stronger cell-cell adhesion. Finally, EEC1,2 provide tissue ductility, which allows cells to undergo higher strains during collective migration, as if cells unjammed from a corticaltensioned state 36 , thereby allowing even greater collective migration speed, (Fig. 7).
Thus, fast collective migration emerges from a fine balance of tissue tensile strength and ductility.
In summary, using a range of microscopy, cell and tissue manipulation assays, we determined that the EEC4 region promotes the recruitment of catenins and vinculin to cell-cell contacts and epithelium-like cohesive cell sheet formation on an extracellular matrix. EEC4 also contributes to the increase in cell cortical tension in an intercellular adhesion-independent manner, likely through a Cdc42 pathway. In turn, the EEC3 is required to provide strong cell-cell SF and, subsequently, tensile strength of the cell sheet, mostly through an increase in dissipation during cell-cell separation rather than an increase in static intercellular adhesion energy. Moreover, EC1,2 together with EEC3 provide an additional increase in cell-cell SF and tensile strength but also antagonize EC4 activity on cell cortical tension for increased ductility to the cell sheet. Finally, we propose that increased ductility compensates the dampening effect of epithelium formation on cell migration speed. Cortical cytoskeleton regulation by the cadherin cytoplasmic region, rather than cadherin trans-interactions, has increasingly appeared as a key determinant of intercellular adhesion energy and cell sorting. Our findings nevertheless remind us of the importance of cadherin extracellular domains in major tissue mechanical properties, and reveal their respective effects and underlying cellular mechanisms. Whether the expression of full-length type-I and type-II cadherins in varying proportions in the same cells throughout development determines the major morphogenetic events through the mechanisms unveiled here is a promising research avenue for future studies.

Compteing interests
The authors declare no competing interests for the work described in this manuscript

Cell lines, constructs, lentivirus-based transductions, and immunostaining
Protein sequences and domain organizations of mouse E-cad and cadherin-7 were used to construct chimeric cadherins, including 77EEE, 777EE, 7777E and EE7EE, as shown in Fig. 1a, b. In these chimeras, we preserved the calcium binding sites.
Sequences were cloned into PLVX-puro vector with eGFP at the C-terminus and proper insertion confirmed by sequencing. HEK293T cells were co-transfected (Effectene transfection reagent) with the plasmid and optimized lentiviral packaging mix (ViraPower TM ). The viral supernatant was harvested, filtered, and added to S180 cells plated at 80% confluence. Stably transduced cells were selected based on puromycin resistance, FACS sorted for a medium level of expression, and checked for mycoplasma contamination (MycoAlert TM , Lonza). Rabbit anti-α-catenin (Sigma-Aldrich), mouse monoclonal anti-β-catenin, mouse anti-p120-catenin (Invitrogen) and mouse monoclonal anti-vinculin (Sigma-Aldrich) were used for immunostaining.
Strand-swap incompetent E-cad (W2A mutant) and X-dimer incompetent E-cad (K14E) were also stably expressed separately in S180 cells following the above lentivirus-based transduction protocol and FACS sorted for medium expression.

Computer model for validation of preserved domain interactions and calcium binding sites in chimeric cadherin ECs
Models of juxtaposed domains at the hybrid interfaces were built using the web-based interface of SWISS-MODEL 34 and existing protein structures as templates: 2A62, 3Q2V, 3Q2W, 1L3W and 5VEB 5,9,47,54 . Calcium ions were placed by hand and sidechains angles were orientated to coordinate the ions in Coot 55 .

Confocal microscopy, image processing and analysis
Confocal images on fixed samples were acquired using a Nikon A1plus microscope.
Z-stacks of cells were acquired from 1 µm below the basal cell surface to 1 µm above the apical cell surface, with a step size of 0.5 µm at 60x magnification. The laser power, pinhole size and other standard parameters were kept identical for all samples. Using the plot profile function, the intensity plot was obtained and the intensity values were listed and saved in Excel sheets for data processing. In a similar manner, data was collected from the cell free area of the coverslip and subtracted to eliminate the background signal. The cell surface cadherin signal intensity per µm length was calculated and normalized to E-cad per µm intensity. In a similar manner, junctional cadherin intensity was quantified by selecting all the slices in the stack for maximum intensity z projection and tracing the cell-cell junctions with the freehand line tool.

Tissue culture, cell dissociation
Cells were maintained in high glucose DMEM with 10% FCS, and confluent cultures were routinely treated with TE buffer (0.05% trypsin, 0.02% EDTA). For force measurements, cells were treated with TE buffer (0.05% trypsin, 0.02% EDTA) and were seeded as single cells in a 60-mm ultra-low adhesion dishes (Corning) at 30,000 cells in 3 ml culture medium. After 18 hrs of suspension culture, cells were gently re-suspended in CO 2 -independent medium (Invitrogen) supplemented with 10% FCS and used immediately for SF measurement experiments.

Measurement of separation force by pillar-pipette assay
A 60-mm ultra-low adhesion dish (Corning) was prepared by cutting a 2-cm window for pipette access. A PDMS block (Sylgard, Dow-Corning) with pillars 35 µm in diameter and 200 µm long arranged in a hexagonal pattern with a pitch of 400 µm was cut into a thin slice of 1-2 mm thickness and 1-cm long. The strip was further cut into 5 pieces of ~2-mm thickness and stuck to the pre-prepared ultra-low adhesion dish in a staggered arrangement so that ~20 micropillars at different x, y, z-levels were available to enable high-throughput. The dish was plasma treated to make the pillars hydrophilic (Harrick Plasma). During plasma treatment, a thin film of PDMS masked the area of the dish where the ultra-low property was to be maintained g. The pillars were coated with 50 µl fibronectin (Sigma-Aldrich) mixed with fluorescently labeled fibronectin (Cytoskeleton) in a ratio of 10:1 for 1 hr at 37 °C. The fibronectin was then washed with deionized water three times and the surface left to dry. The thin film of PDMS was removed before experiments.
Cells were manipulated at 37 °C with a micropipette that was held by a micromanipulator connected to a combined hydraulic/pneumatic system.
Micropipettes were pulled (P-2000, Sutter Instrument) and cut with a microforge instrument (MF-900, Narishige) to an inner diameter of 4-6 µm. Cell doublets were visually observed under bright-field and epifluorescence to identify doublets with mature, bright junctions and a larger contact radius. One end of the doublet was attached to a PDMS pillar tip with the doublet axis parallel to the dish and perpendicular to the pillar's axis (Fig. 2a,b and Supplementary Fig. 3d). After 5 min of attachment, a second pre-existing doublet was selected with the micropipette and brought in contact with the first doublet in a series arrangement (Fig. 2b). The new junction between the two doublets was established over 1 hr. The two doublets were separated from each other using a micropipette while the deflection in the pillar was imaged in bright-field (40× x 1.5, Nikon, Eclipse 2,000) at the rate of 1 image per sec.
The SF required to separate the two doublets (after 1 hr contact) was quantified from the pillar deflection using the following equation (Supplementary Fig. 3d): (1) where L is the effective length of the pillar, D is the diameter, E is the Young's Modulus of the PDMS material, and δ is the maximum pillar deflection before detaching the two doublets from each other (Supplementary Fig. 3d). Upon detaching the doublets, the two cells of the other doublet still attached to the pillar were separated using the micropipette, and the force required to separate the cells (preexisting doublets) was determined. The results from 15 measurements from each group were used to obtain the mean SF for a specific contact time in at least three independent experiments.

Dispase-based junctional protein tensor assay (D-JPT)
Channels made of UV curable polymer (NOA 73, Norland Products) were prepared using a PDMS stencil with the shape of the channel projecting out (100-500 µm high) placed upside-down on the glass-bottomed dish (IWAKI), and the low viscous UV curable polymer made to flow through the channels by capillarity effect and UV cured. The PDMS stencil was peeled off to expose the channel. The channel was washed three times by a jet of deionized water. Cells were seeded at 60% confluence in the rectangular glass-bottomed channel, 1-mm wide and 2.5-cm long with a central gradual constriction of 0.5 mm and 2.5 mm long (Fig. 5a). Cells were seeded within 3 h of making the channels to retain their hydrophobic characteristics to avoid cells attaching to the polymer. Upon reaching >95% confluence (Fig. 5b [ 5d).

Cortical tension measurement and G-LISA assay
Cells were dissociated from confluent cultures by washing once with complete medium to remove loosely attached cells, and then gently flushing a spot of the monolayer to remove a patch of cells. This was done using a 1-ml pipette containing 1 ml CO 2 -independent medium supplemented with 10% FCS and flushing the spot three times with the same 1 ml of medium. The patch of removed cells was observed under the microscope to ensure that there were no cells in the area and thus the isolated cells were not a subpopulation of easily detachable cells with different rheology. The dissociated cells were gently pipetted 5 times to dissociate cell clumps before being deposited on a PLL-PEG-coated (SuSoS), glass-bottomed dish (IWAKI). A 1-cm window for pipette access was cut on the side of the glass-bottomed dish before cell seeding. Micropipettes were pulled as described earlier, and cut to a diameter ranging from 6 to 9 µm. Before each measurement, the pressure in the pipette was equilibrated with the pressure in the dish by monitoring the movement of cells when the pipette was brought closer to the cells. Cortical tension measurements were commenced 30 mins after the deposition of the cells onto the dish; this time delay allowed cells to become spherical at 37 °C. Single cells were gently aspirated into the pipette with a low pressure sufficient to grab the cell. The pressure was then gradually decreased until a threshold pressure was reached, at which point the cell formed a hemispherical protrusion into the pipette equal to the radius of the pipette (Fig. 3a). For measuring the cortical tension of doublets, the tip of the doublets was aspirated (Fig. 3b). The surface tension was computed using the Young-Laplace law γ = P c / (2/R p -2/R c ), where P c is the negative pressure inside the pipette, R c is the radius of the cell and R p the radius of the pipette 30 . Experiments were performed within an environmental chamber maintained at 37 °C.
For G-LISA assay, to quantify active Rac1 and Cdc42 levels, cells were seeded at 50% confluence in a 10 cm cell culture dish. Upon reaching >95% confluence, scratch wounds were created using a 10-µl pipette tip, creating horizontal and vertical wounds at a spacing of ~1 cm. The cultures were washed 3 times with 1x PBS (with calcium and magnesium) to remove the detached cells and incubated for a further 3 h to allow cells to start migrating, a process which is known to be regulated by Rho GTPases.
The cells were then lysed, and G-LISA assay was performed as per the manufacturer's instructions (Rac1 and Cdc42 G-LISA activation assay kits, Cytoskeleton).

Preparation of wound-free gaps for cell migration assay
Wound gap (500 µm) masking inserts (Ibidi) were placed in the middle of the wells of 12-well cell culture-treated plates. Cells were seeded by pipetting 70 µl of cell suspension at 6 × 10 5 cells/ml. After 18-24 h of culture, when the cells had reached >95% confluence, the inserts were carefully removed without disturbing the cells.
The wells were gently washed three times with 1 ml PBS (with calcium and magnesium) to remove loose cells. Fresh medium (1 ml) was added to the culture, and the gap was imaged at different locations to acquire an image every 20 min using an inverted microscope (DMi8, Leica) equipped with an environmental chamber maintained at 37 °C and 5% CO 2 .

Statistical analysis
All statistical analyses were performed with GraphPad Prism 5 software using an unpaired, two-tailed Student's t-test. Data are presented as mean ± SD.          with E-cad, 77EEE and 7777E expressing S180 cells. Note the 77EE7 cells migrate collectively at a speed comparable to 77EEE cells. Also note the EEE7E cells migrating in a less cohesive manner and at speeds comparable to 7777E cells.